You are viewing a plain text version of this content. The canonical link for it is here.

Posted to dev@myriad.apache.org by "DarinJ (JIRA)" <ji...@apache.org> on 2015/10/21 21:22:27 UTC

[jira] [Commented] (MYRIAD-160) Randomize ports on which NM starts

    [ https://issues.apache.org/jira/browse/MYRIAD-160?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14967712#comment-14967712 ] 

DarinJ commented on MYRIAD-160:
-------------------------------

Similar in nature to MYRIAD-155.

It should be easy to get a random port from Mesos (Sample a large enough range), however it may be better solution to remove the Resource Managers memory of the old Node Manager if a node is properly flexed down.

> Randomize ports on which NM starts
> ----------------------------------
>
>                 Key: MYRIAD-160
>                 URL: https://issues.apache.org/jira/browse/MYRIAD-160
>             Project: Myriad
>          Issue Type: Bug
>          Components: Scheduler
>            Reporter: Aashreya Ravi Shankar
>
> When a CGS node is present and gets replaced by a FGS node after a flex-down and flex-up, the memory calculation is going wrong. I see the applications hang and the AM container fails to get scheduled.
> By default when a NM is started, it is started on a random port. Only the web server port is fixed to 8042. 
> The behavior should be similar in Myriad. We need to randomize the port of the NM from the offer obtained from mesos. So when we flex down and flex-up it starts on a different port. Otherwise Myriad considers this as the same NM with different resources. Hence the memory calculation issue. 



--
This message was sent by Atlassian JIRA
(v6.3.4#6332)